Abstracts

Rationale: One of the most crippling aspects of epilepsy is the unpredictable nature of seizures. Despite intense efforts, seizure prediction remains empirical and insufficiently accurate for clinical use. We used an in-silico model of post-traumatic epileptogenesis that displays both interictal and ictal activity to test whether there are network states that precede seizures.Methods: We modeled network activity before and after trauma during different stages of epileptogenesis using a large scale computational model of hippocampal area. The model was as a matrix of 105 x 105 integrate-and-fire neurons and an 18 x 18 array of interneurons. Cells were synaptically connected via glutamatergic and GABAergic connections that were stochastically generated using the same connectivity probabilities. Fractional cell loss due to trauma was equally and randomly distributed in both glutamatergic and GABAergic networks. Traumatic cell loss was compensated by subsequent axon sprouting to re-constitute the original number of synapses.Results: After trauma and sprouting, the recurrent connectivity in the network led to bursts of synchronous activity. Models in which long-distance sprouting was permitted demonstrated an increased in frequency of these burstst. Long-distance sprouting also increased the fraction of neurons that participated in the synchronous bursts. With sufficient levels of cell loss and sprouting distances, the network also produced sustained (tonic) ictal-like activity that slowly transitioned into intermittent (clonic) activity. To understand the necessary conditions for seizures, we analyzed spatial and temporal patterns of synchronous activity propagation, synaptic depression and recovery, and local and global imbalances between inhibition and excitation. We found that recurrent sprouting was necessary but not sufficient for development of ictal-like activity. We found that the critical determinant of ictal transitions was the amount of variance in the degree of activity-dependent synaptic depression. We also found that transition from tonic to clonic activity and eventually back to periodic synchronous bursting was not caused by global activity-dependent synaptic depression, but depended instead on the anatomical distribution of synaptic depression and recovery. Those differences were relatively small and it was surprisingly difficult to correlate them with obvious changes in network activity.Conclusions: Our results suggest that in order to understand ictogenesis we need to develop methods to accurately estimate the degree of recurrent connectivity in the network but also to measure subtle, activity-dependent inhomogeneities permissive for seizure onset. This will require detailed knowledge of both the anatomy and functional states of recurrent synaptic connectivity in epileptic networks. In theory this information would be available from activity-dependent imaging at sufficiently high temporal and spatial resolution.